glibc/FAQ.in

	    Frequently Asked Questions about the GNU C Library

This document tries to answer questions a user might have when installing
and using glibc.  Please make sure you read this before sending questions or
bug reports to the maintainers.

The GNU C library is very complex.  The installation process has not been
completely automated; there are too many variables.  You can do substantial
damage to your system by installing the library incorrectly.  Make sure you
understand what you are undertaking before you begin.

If you have any questions you think should be answered in this document,
please let me know.

						  --drepper@cygnus.com

? Compiling glibc

??	What systems does the GNU C Library run on?

{UD} This is difficult to answer.  The file `README' lists the architectures
GNU libc was known to run on *at some time*.  This does not mean that it
still can be compiled and run on them now.

The systems glibc is known to work on as of this release, and most probably
in the future, are:

	*-*-gnu			GNU Hurd
	i[3456]86-*-linux-gnu	Linux-2.x on Intel
	m68k-*-linux-gnu	Linux-2.x on Motorola 680x0
	alpha-*-linux-gnu	Linux-2.x on DEC Alpha
	powerpc-*-linux-gnu     Linux and MkLinux on PowerPC systems
	sparc-*-linux-gnu	Linux-2.x on SPARC
	sparc64-*-linux-gnu	Linux-2.x on UltraSPARC
	arm-*-none		ARM standalone systems
	arm-*-linux		Linux-2.x on ARM
	arm-*-linuxaout		Linux-2.x on ARM using a.out binaries

Ports to other Linux platforms are in development, and may in fact work
already, but no one has sent us success reports for them.  Currently no
ports to other operating systems are underway, although a few people have
expressed interest.

If you have a system not listed above (or in the `README' file) and you are
really interested in porting it, contact

	<bug-glibc@gnu.org>

??	What compiler do I need to build GNU libc?

{UD} You must use GNU CC to compile GNU libc.  A lot of extensions of GNU CC
are used to increase portability and speed.

GNU CC is found, like all other GNU packages, on

	ftp://ftp.gnu.org/pub/gnu

and the many mirror sites.  ftp.gnu.org is always overloaded, so try to find
a local mirror first.

You should always try to use the latest official release.  Older versions
may not have all the features GNU libc requires.  The current releases of
egcs (1.0.3 and 1.1) and GNU CC (2.8.1) should work with the GNU C library
(for powerpc see question ?powerpc; for ARM see question ?arm).

{ZW} You may have problems if you try to mix code compiled with
EGCS and with GCC 2.8.1. See ?exception for details.

??	When I try to compile glibc I get only error messages.
	What's wrong?

{UD} You definitely need GNU make to translate GNU libc.  No other make
program has the needed functionality.

We recommend version GNU make version 3.75 or 3.77.  Versions before 3.75
have bugs and/or are missing features.  Version 3.76 has bugs which
appear when building big projects like GNU libc. 3.76.1 appears to work but
some people have reported problems.

??	Do I need a special linker or assembler?

{ZW} If you want a shared library, you need a linker and assembler that
understand all the features of ELF, including weak and versioned symbols.
The static library can be compiled with less featureful tools, but lacks key
features such as NSS.

For Linux or Hurd, you want binutils 2.8.1.0.23, 2.9.1, or 2.9.1.0.15 or
higher.  These are the only versions we've tested and found reliable.  Other
versions after 2.8.1.0.23 may work but we don't recommend them, especially
not when C++ is involved.  Earlier versions do not work at all.

Other operating systems may come with system tools that have all the
necessary features, but this is moot because glibc hasn't been ported to
them.

??powerpc	Which compiler should I use for powerpc?

{GK} You want to use egcs 1.1 or later (together with the right versions
of all the other tools, of course).

In fact, egcs 1.1 has a bug that causes linuxthreads to be
miscompiled, resulting in segmentation faults when using condition
variables.  There is a temporary patch at:

<http://discus.anu.edu.au/~geoffk/egcs-3.diff>

Later versions of egcs may fix this problem.

??arm	Which tools should I use for ARM?

{PB} You should use egcs 1.1 or a later version.  For ELF systems some
changes are needed to the compiler; a patch against egcs-1.1.x can be found
at:

<ftp://ftp.netwinder.org/users/p/philb/egcs-1.1.1pre2-diff-981126>

Binutils 2.9.1.0.16 or later is also required.

??	Do I need some more things to compile the GNU C Library?

{UD} Yes, there are some more :-).

* GNU gettext.  This package contains the tools needed to construct
  `message catalog' files containing translated versions of system
  messages. See ftp://ftp.gnu.org/pub/gnu or better any mirror
  site.  (We distribute compiled message catalogs, but they may not be
  updated in patches.)

* Some files are built with special tools.  E.g., files ending in .gperf
  need a `gperf' program.  The GNU version (now available in a separate
  package, formerly only as part of libg++) is known to work while some
  vendor versions do not.

  You should not need these tools unless you change the source files.

* Perl 5 is needed if you wish to test an installation of GNU libc
  as the primary C library.

* When compiling for Linux, the header files of the Linux kernel must
  be available to the compiler as <linux/*.h> and <asm/*.h>.

* lots of disk space (~170MB for i?86-linux; more for RISC platforms,
  as much as 400MB).

* plenty of time.  Compiling just the shared and static libraries for
  i?86-linux takes approximately 1h on an i586@133, or 2.5h on
  i486@66, or 4.5h on i486@33.  Multiply this by 1.5 or 2.0 if you
  build profiling and/or the highly optimized version as well.  For
  Hurd systems times are much higher.

  You should avoid compiling in a NFS mounted filesystem.  This is
  very slow.

  James Troup <J.J.Troup@comp.brad.ac.uk> reports a compile time of
  45h34m for a full build (shared, static, and profiled) on Atari
  Falcon (Motorola 68030 @ 16 Mhz, 14 Mb memory) and Jan Barte
  <yann@plato.uni-paderborn.de> reports 22h48m on Atari TT030
  (Motorola 68030 @ 32 Mhz, 34 Mb memory)

  If you have some more measurements let me know.

??	What version of the Linux kernel headers should be used?

{AJ,UD} The headers from the most recent Linux kernel should be used.  The
headers used while compiling the GNU C library and the kernel binary used
when using the library do not need to match.  The GNU C library runs without
problems on kernels that are older than the kernel headers used.  The other
way round (compiling the GNU C library with old kernel headers and running
on a recent kernel) does not necessarily work.  For example you can't use
new kernel features when using old kernel headers for compiling the GNU C
library.

{ZW} Even if you are using a 2.0 kernel on your machine, we recommend you
compile GNU libc with 2.1 kernel headers.  That way you won't have to
recompile libc if you ever upgrade to kernel 2.1 or 2.2.  To tell libc which
headers to use, give configure the --with-headers switch
(e.g. --with-headers=/usr/src/linux-2.1.107/include).

Note that you must configure the 2.1 kernel if you do this, otherwise libc
will be unable to find <linux/version.h>.  Just copy .config from your 2.0
kernel sources to the 2.1 tree, do `make oldconfig', and say no to all the
new options.

??	The compiler hangs while building iconvdata modules.  What's
	wrong?

{ZW} This is a problem with old versions of GCC.  Initialization of large
static arrays is very slow.  The compiler will eventually finish; give it
time.

The problem is fixed in egcs 1.1 but not in earlier releases.

??	When I run `nm -u libc.so' on the produced library I still
	find unresolved symbols.  Can this be ok?

{UD} Yes, this is ok.  There can be several kinds of unresolved symbols:

* magic symbols automatically generated by the linker.  These have names
  like __start_* and __stop_*

* symbols starting with _dl_* come from the dynamic linker

* weak symbols, which need not be resolved at all (fabs for example)

Generally, you should make sure you find a real program which produces
errors while linking before deciding there is a problem.

??addon	What are these `add-ons'?

{UD} To avoid complications with export rules or external source code some
optional parts of the libc are distributed as separate packages (e.g., the
crypt package, see ?crypt).

To use these packages as part of GNU libc, just unpack the tarfiles in the
libc source directory and tell the configuration script about them using the
--enable-add-ons option.  If you give just --enable-add-ons configure tries
to find all the add-on packages in your source tree.  This may not work.  If
it doesn't, or if you want to select only a subset of the add-ons, give a
comma-separated list of the add-ons to enable:

	configure --enable-add-ons=crypt,linuxthreads

for example.

Add-ons can add features (including entirely new shared libraries), override
files, provide support for additional architectures, and just about anything
else.  The existing makefiles do most of the work; only some few stub rules
must be written to get everything running.

??	My XXX kernel emulates a floating-point coprocessor for me.
	Should I enable --with-fp?

{ZW} An emulated FPU is just as good as a real one, as far as the C library
is concerned.  You only need to say --without-fp if your machine has no way
to execute floating-point instructions.

People who are interested in squeezing the last drop of performance
out of their machine may wish to avoid the trap overhead, but this is
far more trouble than it's worth: you then have to compile
*everything* this way, including the compiler's internal libraries
(libgcc.a for GNU C), because the calling conventions change.

??	When compiling GNU libc I get lots of errors saying functions
	in glibc are duplicated in libgcc.

{EY} This is *exactly* the same problem that I was having.  The problem was
due to the fact that configure didn't correctly detect that the linker flag
--no-whole-archive was supported in my linker.  In my case it was because I
had run ./configure with bogus CFLAGS, and the test failed.

One thing that is particularly annoying about this problem is that once this
is misdetected, running configure again won't fix it unless you first delete
config.cache.

{UD} Starting with glibc-2.0.3 there should be a better test to avoid some
problems of this kind.  The setting of CFLAGS is checked at the very
beginning and if it is not usable `configure' will bark.

??	Why do I get messages about missing thread functions when I use
	librt?  I don't even use threads.

{UD} In this case you probably mixed up your installation.  librt uses
threads internally and has implicit references to the thread library.
Normally these references are satisfied automatically but if the thread
library is not in the expected place you must tell the linker where it is.
When using GNU ld it works like this:

	gcc -o foo foo.c -Wl,-rpath-link=/some/other/dir -lrt

The `/some/other/dir' should contain the thread library.  `ld' will use the
given path to find the implicitly referenced library while not disturbing
any other link path.

??	What's the problem with configure --enable-omitfp?

{AJ} When --enable-omitfp is set the libraries are built without frame
pointers.  Some compilers produce buggy code for this model and therefore we
don't advise using it at the moment.

If you use --enable-omitfp, you're on your own.  If you encounter problems
with a library that was build this way, we advise you to rebuild the library
without --enable-omitfp.  If the problem vanishes consider tracking the
problem down and report it as compiler failure.

Since a library build with --enable-omitfp is undebuggable on most systems,
debuggable libraries are also built - you can use it by appending "_g" to
the library names.

The compilation of these extra libraries and the compiler optimizations slow
down the build process and need more disk space.

??	I get failures during `make check'.  What shall I do?

{AJ} The testsuite should compile and run cleanly on your system, every
failure should be looked into.  Depending on the failure I wouldn't advise
installing the library at all.

You should consider using the `glibcbug' script to report the failure,
providing as much detail as possible.  If you run a test directly, please
remember to set up the environment correctly.  You want to test the compiled
library - and not your installed one.  The best way is to copy the exact
command line which failed and run the test from the subdirectory for this
test in the sources.

There are some failures which are not directly related to the GNU libc:
- Some compiler produce buggy code.  The egcs 1.1 release should be ok.  gcc
  2.8.1 might cause some failures, gcc 2.7.2.x is so buggy, that explicit
  checks have been used so that you can't build with it.
- The kernel might have bugs.  For example on Linux/Alpha 2.0.34 the
  floating point handling has quite a number of bugs and therefore most of
  the test cases in the math subdirectory will fail.  The current Linux 2.1
  development kernels have fixes for the floating point support on Alpha.

??	What is symbol versioning good for?  Do I need it?

{AJ} Symbol versioning solves problems that are related to interface
changes.  One version of an interface might have been introduced in a
previous version of the GNU C library but the interface or the semantics of
the function has been changed in the meantime.  For binary compatibility
with the old library, a newer library needs to still have the old interface
for old programs.  On the other hand new programs should use the new
interface.  Symbol versioning is the solution for this problem.  The GNU
libc version 2.1 uses by default symbol versioning if the binutils support
it.

We don't advise to build without symbol versioning since you lose binary
compatibility if you do - for ever!  The binary compatibility you lose is
not only against the previous version of the GNU libc (version 2.0) but also
against future versions.


? Installation and configuration issues

??	Can I replace the libc on my Linux system with GNU libc?

{UD} You cannot replace any existing libc for Linux with GNU libc.  It is
binary incompatible and therefore has a different major version.  You can,
however, install it alongside your existing libc.

For Linux there are three major libc versions:
	libc-4		a.out libc
	libc-5		original ELF libc
	libc-6		GNU libc

You can have any combination of these three installed.  For more information
consult documentation for shared library handling.  The Makefiles of GNU
libc will automatically generate the needed symbolic links which the linker
will use.

??	How do I configure GNU libc so that the essential libraries
	like libc.so go into /lib and the other into /usr/lib?

{UD,AJ} Like all other GNU packages GNU libc is designed to use a base
directory and install all files relative to this.  The default is
/usr/local, because this is safe (it will not damage the system if installed
there).  If you wish to install GNU libc as the primary C library on your
system, set the base directory to /usr (i.e. run configure --prefix=/usr
<other_options>).  Note that this can damage your system; see ?safety for
details.

Some systems like Linux have a filesystem standard which makes a difference
between essential libraries and others.  Essential libraries are placed in
/lib because this directory is required to be located on the same disk
partition as /.  The /usr subtree might be found on another
partition/disk. If you configure for Linux with --prefix=/usr, then this
will be done automatically.

To install the essential libraries which come with GNU libc in /lib on
systems other than Linux one must explicitly request it.  Autoconf has no
option for this so you have to use a `configparms' file (see the `INSTALL'
file for details).  It should contain:

slibdir=/lib
sysconfdir=/etc

The first line specifies the directory for the essential libraries, the
second line the directory for system configuration files.

??safety	How should I avoid damaging my system when I install GNU libc?

{ZW} If you wish to be cautious, do not configure with --prefix=/usr.  If
you don't specify a prefix, glibc will be installed in /usr/local, where it
will probably not break anything.  (If you wish to be certain, set the
prefix to something like /usr/local/glibc2 which is not used for anything.)

The dangers when installing glibc in /usr are twofold:

* glibc will overwrite the headers in /usr/include.  Other C libraries
  install a different but overlapping set of headers there, so the
  effect will probably be that you can't compile anything.  You need to
  rename /usr/include out of the way first.  (Do not throw it away; you
  will then lose the ability to compile programs against your old libc.)

* None of your old libraries, static or shared, can be used with a
  different C library major version.  For shared libraries this is not a
  problem, because the filenames are different and the dynamic linker
  will enforce the restriction.  But static libraries have no version
  information.  You have to evacuate all the static libraries in
  /usr/lib to a safe location.

The situation is rather similar to the move from a.out to ELF which
long-time Linux users will remember.

??	Do I need to use GNU CC to compile programs that will use the
	GNU C Library?

{ZW} In theory, no; the linker does not care, and the headers are supposed
to check for GNU CC before using its extensions to the C language.

However, there are currently no ports of glibc to systems where another
compiler is the default, so no one has tested the headers extensively
against another compiler.  You may therefore encounter difficulties.  If you
do, please report them as bugs.

Also, in several places GNU extensions provide large benefits in code
quality.  For example, the library has hand-optimized, inline assembly
versions of some string functions.  These can only be used with GCC.  See
?string for details.

??crypt	When linking with the new libc I get unresolved symbols
	`crypt' and `setkey'.  Why aren't these functions in the
	libc anymore?

{UD} The US places restrictions on exporting cryptographic programs and
source code.  Until this law gets abolished we cannot ship the cryptographic
functions together with glibc.

The functions are available, as an add-on (see ?addon).  People in the US
may get it from the same place they got GNU libc from.  People outside the
US should get the code from ftp://ftp.ifi.uio.no/pub/gnu, or another archive
site outside the USA.  The README explains how to install the sources.

If you already have the crypt code on your system the reason for the failure
is probably that you did not link with -lcrypt.  The crypto functions are in
a separate library to make it possible to export GNU libc binaries from the
US.

??	When I use GNU libc on my Linux system by linking against
	the libc.so which comes with glibc all I get is a core dump.

{UD} On Linux, gcc sets the dynamic linker to /lib/ld-linux.so.1 unless the
user specifies a -dynamic-linker argument.  This is the name of the libc5
dynamic linker, which does not work with glibc.

For casual use of GNU libc you can just specify to the linker
    --dynamic-linker=/lib/ld-linux.so.2

which is the glibc dynamic linker, on Linux systems.  On other systems the
name is /lib/ld.so.1.  When linking via gcc, you've got to add
    -Wl,--dynamic-linker=/lib/ld-linux.so.2

to the gcc command line.

To change your environment to use GNU libc for compiling you need to change
the `specs' file of your gcc.  This file is normally found at

	/usr/lib/gcc-lib/<arch>/<version>/specs

In this file you have to change a few things:

- change `ld-linux.so.1' to `ld-linux.so.2'

- remove all expression `%{...:-lgmon}';  there is no libgmon in glibc

- fix a minor bug by changing %{pipe:-} to %|

Here is what the gcc-2.7.2 specs file should look like when GNU libc is
installed at /usr:

-----------------------------------------------------------------------
*asm:
%{V} %{v:%{!V:-V}} %{Qy:} %{!Qn:-Qy} %{n} %{T} %{Ym,*} %{Yd,*} %{Wa,*:%*}

*asm_final:
%|

*cpp:
%{fPIC:-D__PIC__ -D__pic__} %{fpic:-D__PIC__ -D__pic__} %{!m386:-D__i486__} %{posix:-D_POSIX_SOURCE} %{pthread:-D_REENTRANT}

*cc1:
%{profile:-p}

*cc1plus:


*endfile:
%{!shared:crtend.o%s} %{shared:crtendS.o%s} crtn.o%s

*link:
-m elf_i386 %{shared:-shared}   %{!shared:     %{!ibcs:       %{!static: 	%{rdynamic:-export-dynamic} 	%{!dynamic-linker:-dynamic-linker /lib/ld-linux.so.2}} 	%{static:-static}}}

*lib:
%{!shared: %{pthread:-lpthread} 	%{profile:-lc_p} %{!profile: -lc}}

*libgcc:
-lgcc

*startfile:
%{!shared:      %{pg:gcrt1.o%s} %{!pg:%{p:gcrt1.o%s} 		     %{!p:%{profile:gcrt1.o%s} 			 %{!profile:crt1.o%s}}}}    crti.o%s %{!shared:crtbegin.o%s} %{shared:crtbeginS.o%s}

*switches_need_spaces:


*signed_char:
%{funsigned-char:-D__CHAR_UNSIGNED__}

*predefines:
-D__ELF__ -Dunix -Di386 -Dlinux -Asystem(unix) -Asystem(posix) -Acpu(i386) -Amachine(i386)

*cross_compile:
0

*multilib:
. ;

-----------------------------------------------------------------------

Things get a bit more complicated if you have GNU libc installed in some
other place than /usr, i.e., if you do not want to use it instead of the old
libc.  In this case the needed startup files and libraries are not found in
the regular places.  So the specs file must tell the compiler and linker
exactly what to use.

Version 2.7.2.3 does and future versions of GCC will automatically
provide the correct specs.

??	Looking through the shared libc file I haven't found the
	functions `stat', `lstat', `fstat', and `mknod' and while
	linking on my Linux system I get error messages.  How is
	this supposed to work?

{RM} Believe it or not, stat and lstat (and fstat, and mknod) are supposed
to be undefined references in libc.so.6!  Your problem is probably a missing
or incorrect /usr/lib/libc.so file; note that this is a small text file now,
not a symlink to libc.so.6.  It should look something like this:

GROUP ( libc.so.6 libc_nonshared.a )

??exception	When I run an executable on one system which I compiled on
	another, I get dynamic linker errors.  Both systems have the same
	version of glibc installed.  What's wrong?

{ZW} Glibc on one of these systems was compiled with gcc 2.7 or 2.8, the
other with egcs (any version).  Egcs has functions in its internal
`libgcc.a' to support exception handling with C++.  They are linked into
any program or dynamic library compiled with egcs, whether it needs them or
not.  Dynamic libraries then turn around and export those functions again
unless special steps are taken to prevent them.

When you link your program, it resolves its references to the exception
functions to the ones exported accidentally by libc.so.  That works fine as
long as libc has those functions.  On the other system, libc doesn't have
those functions because it was compiled by gcc 2.8, and you get undefined
symbol errors.  The symbols in question are named things like
`__register_frame_info'.

For glibc 2.0, the workaround is to not compile libc with egcs.  We've also
incorporated a patch which should prevent the EH functions sneaking into
libc.  It doesn't matter what compiler you use to compile your program.

For glibc 2.1, we've chosen to do it the other way around: libc.so
explicitly provides the EH functions.  This is to prevent other shared
libraries from doing it.  You must therefore compile glibc 2.1 with EGCS
unless you don't care about ever importing binaries from other systems.
Again, it doesn't matter what compiler you use for your programs.

??	How can I compile gcc 2.7.2.1 from the gcc source code using
	glibc 2.x?

{AJ} There's only correct support for glibc 2.0.x in gcc 2.7.2.3 or later.
But you should get at least gcc 2.8.1 or egcs 1.0.2 (or later versions)
instead.

??	The `gencat' utility cannot process the catalog sources which
	were used on my Linux libc5 based system.  Why?

{UD} The `gencat' utility provided with glibc complies to the XPG standard.
The older Linux version did not obey the standard, so they are not
compatible.

To ease the transition from the Linux version some of the non-standard
features are also present in the `gencat' program of GNU libc.  This mainly
includes the use of symbols for the message number and the automatic
generation of header files which contain the needed #defines to map the
symbols to integers.

Here is a simple SED script to convert at least some Linux specific catalog
files to the XPG4 form:

-----------------------------------------------------------------------
# Change catalog source in Linux specific format to standard XPG format.
# Ulrich Drepper <drepper@cygnus.com>, 1996.
#
/^\$ #/ {
  h
  s/\$ #\([^ ]*\).*/\1/
  x
  s/\$ #[^ ]* *\(.*\)/\$ \1/
}

/^# / {
  s/^# \(.*\)/\1/
  G
  s/\(.*\)\n\(.*\)/\2 \1/
}
-----------------------------------------------------------------------

??	Programs using libc have their messages translated, but other
	behavior is not localized (e.g. collating order); why?

{ZW} Translated messages are automatically installed, but the locale
database that controls other behaviors is not.  You need to run localedef to
install this database, after you have run `make install'.  For example, to
set up the French Canadian locale, simply issue the command

    localedef -i fr_CA -f ISO-8859-1 fr_CA

Please see localedata/README in the source tree for further details.

??	I have set up /etc/nis.conf, and the Linux libc 5 with NYS
	works great.  But the glibc NIS+ doesn't seem to work.

{TK} The glibc NIS+ implementation uses a /var/nis/NIS_COLD_START file for
storing information about the NIS+ server and their public keys, because the
nis.conf file does not contain all the necessary information.  You have to
copy a NIS_COLD_START file from a Solaris client (the NIS_COLD_START file is
byte order independent) or generate it with nisinit from the nis-tools
package; available at

    http://www-vt.uni-paderborn.de/~kukuk/linux/nisplus.html

??	I have killed ypbind to stop using NIS, but glibc
	continues using NIS.

{TK} For faster NIS lookups, glibc uses the /var/yp/binding/ files from
ypbind.  ypbind 3.3 and older versions don't always remove these files, so
glibc will continue to use them.  Other BSD versions seem to work correctly.
Until ypbind 3.4 is released, you can find a patch at

    ftp://ftp.kernel.org/pub/linux/utils/net/NIS/ypbind-3.3-glibc4.diff.gz

??	Under Linux/Alpha, I always get "do_ypcall: clnt_call:
	RPC: Unable to receive; errno = Connection refused" when using NIS.

{TK} You need a ypbind version which is 64bit clean.  Some versions are not
64bit clean.  A 64bit clean implementation is ypbind-mt.  For ypbind 3.3,
you need the patch from ftp.kernel.org (See the previous question).  I don't
know about other versions.


??	After installing glibc name resolving doesn't work properly.

{AJ} You probably should read the manual section describing nsswitch.conf
(just type `info libc "NSS Configuration File"').  The NSS configuration
file is usually the culprit.


??	How do I create the databases for NSS?

{AJ} If you have an entry "db" in /etc/nsswitch.conf you should also create
the database files.  The glibc sources contain a Makefile which does the
necessary conversion and calls to create those files.  The file is
`db-Makefile' in the subdirectory `nss' and you can call it with `make -f
db-Makefile'.  Please note that not all services are capable of using a
database.  Currently passwd, group, ethers, protocol, rpc, services shadow
and netgroup are implemented.

??	I have /usr/include/net and /usr/include/scsi as symlinks
	into my Linux source tree.  Is that wrong?

{PB} This was necessary for libc5, but is not correct when using glibc.
Including the kernel header files directly in user programs usually does not
work (see ?kerhdr).  glibc provides its own <net/*> and <scsi/*> header
files to replace them, and you may have to remove any symlink that you have
in place before you install glibc.  However, /usr/include/asm and
/usr/include/linux should remain as they were.

??	Programs like `logname', `top', `uptime' `users', `w' and
	`who', show incorrect information about the (number of)
	users on my system.  Why?

{MK} See ?getlog.

??	After upgrading to glibc 2.1 with symbol versioning I get
	errors about undefined symbols.  What went wrong?

{AJ} The problem is caused either by wrong program code or tools.  In the
versioned libc a lot of symbols are now local that were global symbols in
previous versions.  It seems that programs linked against older versions
often accidentally used libc global variables -- something that should not
happen.

The only way to fix this is to recompile your program. Sorry, that's the
price you might have to pay once for quite a number of advantages with
symbol versioning.

??	When I start the program XXX after upgrading the library
	I get
	  XXX: Symbol `_sys_errlist' has different size in shared
	  object, consider re-linking
	Why?  What should I do?

{UD} As the message says, relink the binary.  The problem is that a few
symbols from the library can change in size and there is no way to avoid
this.  _sys_errlist is a good example.  Occasionally there are new error
numbers added to the kernel and this must be reflected at user level,
breaking programs that refer to them directly.

Such symbols should normally not be used at all.  There are mechanisms to
avoid using them.  In the case of _sys_errlist, there is the strerror()
function which should _always_ be used instead.  So the correct fix is to
rewrite that part of the application.

In some situations (especially when testing a new library release) it might
be possible that a symbol changed size when that should not have happened.
So in case of doubt report such a warning message as a problem.

??	What do I need for C++ development?

{HJ,AJ} You need either egcs 1.1 which comes directly with libstdc++ or
gcc-2.8.1 together with libstdc++ 2.8.1.1.  egcs 1.1 has the better C++
support and works directly with glibc 2.1.  If you use gcc-2.8.1 with
libstdc++ 2.8.1.1, you need to modify libstdc++ a bit.  A patch is available
as:
	ftp://alpha.gnu.org/gnu/libstdc++-2.8.1.1-glibc2.1-diff.gz

Please note that libg++ 2.7.2 (and the Linux Versions 2.7.2.x) doesn't work
very well with the GNU C library due to vtable thunks.  If you're upgrading
from glibc 2.0.x to 2.1 you have to recompile libstdc++ since the library
compiled for 2.0 is not compatible due to the new Large File Support (LFS)
in version 2.1.

{UD} But since in the case of a shared libstdc++ the version numbers should
be different existing programs will continue to work.

??	Even statically linked programs need some shared libraries
	which is not acceptable for me.  What can I do?

{AJ} NSS (for details just type `info libc "Name Service Switch"') won't
work properly without shared libraries.  NSS allows using different services
(e.g. NIS, files, db, hesiod) by just changing one configuration file
(/etc/nsswitch.conf) without relinking any programs.  The only disadvantage
is that now static libraries need to access shared libraries.  This is
handled transparently by the GNU C library.

A solution is to configure glibc with --enable-static-nss.  In this case you
can create a static binary that will use only the services dns and files
(change /etc/nsswitch.conf for this).  You need to link explicitly against
all these services. For example:

  gcc -static test-netdb.c -o test-netdb.c \
    -lc -lnss_files -lnss_dns -lresolv

The problem with this approach is that you've got to link every static
program that uses NSS routines with all those libraries.

{UD} In fact, one cannot say anymore that a libc compiled with this
option is using NSS.  There is no switch anymore.  Therefore it is
*highly* recommended *not* to use --enable-static-nss since this makes
the behaviour of the programs on the system inconsistent.

??	I just upgraded my Linux system to glibc and now I get
	errors whenever I try to link any program.

{ZW} This happens when you have installed glibc as the primary C library but
have stray symbolic links pointing at your old C library.  If the first
`libc.so' the linker finds is libc 5, it will use that.  Your program
expects to be linked with glibc, so the link fails.

The most common case is that glibc put its `libc.so' in /usr/lib, but there
was a `libc.so' from libc 5 in /lib, which gets searched first.  To fix the
problem, just delete /lib/libc.so.  You may also need to delete other
symbolic links in /lib, such as /lib/libm.so if it points to libm.so.5.

{AJ} The perl script test-installation.pl which is run as last step during
an installation of glibc that is configured with --prefix=/usr should help
detect these situations.  If the script reports problems, something is
really screwed up.

??	When I use nscd the machine freezes.

{UD} You cannot use nscd with Linux 2.0.*.  There is functionality missing
in the kernel and work-arounds are not suitable.  Besides, some parts of the
kernel are too buggy when it comes to using threads.

If you need nscd, you have to use a 2.1 kernel.

Note that I have at this point no information about any other platform.

??	I need lots of open files.  What do I have to do?

{AJ} This is at first a kernel issue.  The kernel defines limits with
OPEN_MAX the number of simultaneous open files and with FD_SETSIZE the
number of used file descriptors.  You need to change these values in your
kernel and recompile the kernel so that the kernel allows to use more open
files.  You don't necessarily need to recompile the GNU C library since the
only place where OPEN_MAX and FD_SETSIZE is really needed in the library
itself is the size of fd_set which is used by select.

The GNU C library is now (nearly) select free.  This means it internally has
no limits imposed by the `fd_set' type.  Instead almost all places where the
functionality is needed the `poll' function is used.

If you increase the number of file descriptors in the kernel you don't need
to recompile the C library.  The remaining select calls are in the RPC code.
If your RPC daemons don't need more than FD_SETSIZE file descriptors, you
don't need to change anything at all.

{UD} You can always get the maximum number of file descriptors a process is
allowed to have open at any time using

	number = sysconf (_SC_OPEN_MAX);

This will work even if the kernel limits change.


? Source and binary incompatibilities, and what to do about them

??	I expect GNU libc to be 100% source code compatible with
	the old Linux based GNU libc.  Why isn't it like this?

{DMT,UD} Not every extension in Linux libc's history was well thought-out.
In fact it had a lot of problems with standards compliance and with
cleanliness.  With the introduction of a new version number these errors can
now be corrected.  Here is a list of the known source code
incompatibilities:

* _GNU_SOURCE: glibc does not make the GNU extensions available
  automatically.  If a program depends on GNU extensions or some
  other non-standard functionality, it is necessary to compile it
  with the C compiler option -D_GNU_SOURCE, or better, to put
  `#define _GNU_SOURCE' at the beginning of your source files, before
  any C library header files are included.  This difference normally
  manifests itself in the form of missing prototypes and/or data type
  definitions.  Thus, if you get such errors, the first thing you
  should do is try defining _GNU_SOURCE and see if that makes the
  problem go away.

  For more information consult the file `NOTES' in the GNU C library
  sources.

* reboot(): GNU libc sanitizes the interface of reboot() to be more
  compatible with the interface used on other OSes.  reboot() as
  implemented in glibc takes just one argument.  This argument
  corresponds to the third argument of the Linux reboot system call.
  That is, a call of the form reboot(a, b, c) needs to be changed into
  reboot(c).  Beside this the header <sys/reboot.h> defines the needed
  constants for the argument.  These RB_* constants should be used
  instead of the cryptic magic numbers.

* swapon(): the interface of this function didn't change, but the
  prototype is in a separate header file <sys/swap.h>.  This header
  file also provides the SWAP_* constants defined by <linux/swap.h>;
  you should use them for the second argument to swapon().

* errno: If a program uses the variable "errno", then it _must_
  include <errno.h>.  The old libc often (erroneously) declared this
  variable implicitly as a side-effect of including other libc header
  files.  glibc is careful to avoid such namespace pollution, which,
  in turn, means that you really need to include the header files that
  you depend on.  This difference normally manifests itself in the
  form of the compiler complaining about references to an undeclared
  symbol "errno".

* Linux-specific syscalls: All Linux system calls now have appropriate
  library wrappers and corresponding declarations in various header files.
  This is because the syscall() macro that was traditionally used to
  work around missing syscall wrappers are inherently non-portable and
  error-prone.  The following table lists all the new syscall stubs,
  the header-file declaring their interface and the system call name.

       syscall name:	wrapper name:	declaring header file:
       -------------	-------------	----------------------
       bdflush		bdflush		<sys/kdaemon.h>
       syslog		ksyslog_ctl	<sys/klog.h>

* lpd: Older versions of lpd depend on a routine called _validuser().
  The library does not provide this function, but instead provides
  __ivaliduser() which has a slightly different interface.  Simply
  upgrading to a newer lpd should fix this problem (e.g., the 4.4BSD
  lpd is known to be working).

* resolver functions/BIND: like on many other systems the functions of
  the resolver library are not included in libc itself.  There is a
  separate library libresolv.  If you get undefined symbol errors for
  symbols starting with `res_*' simply add -lresolv to your linker
  command line.

* the `signal' function's behavior corresponds to the BSD semantic and
  not the SysV semantic as it was in libc-5.  The interface on all GNU
  systems shall be the same and BSD is the semantic of choice.  To use
  the SysV behavior simply use `sysv_signal', or define _XOPEN_SOURCE.
  See ?signal for details.

??getlog	Why does getlogin() always return NULL on my Linux box?

{UD} The GNU C library has a format for the UTMP and WTMP file which differs
from what your system currently has.  It was extended to fulfill the needs
of the next years when IPv6 is introduced.  The record size is different and
some fields have different positions.  The files written by functions from
the one library cannot be read by functions from the other library.  Sorry,
but this is what a major release is for.  It's better to have a cut now than
having no means to support the new techniques later.

{MK} There is however a (partial) solution for this problem.  Please take a
look at the file `login/README.utmpd'.

??	Where are the DST_* constants found in <sys/time.h> on many
	systems?

{UD} These constants come from the old BSD days and are not used anymore
(libc5 does not actually implement the handling although the constants are
defined).

Instead GNU libc contains zone database support and compatibility code for
POSIX TZ environment variable handling.

??	The prototypes for `connect', `accept', `getsockopt',
	`setsockopt', `getsockname', `getpeername', `send',
	`sendto', and `recvfrom' are different in GNU libc from
	any other system I saw.  This is a bug, isn't it?

{UD} No, this is no bug.  This version of GNU libc already follows the new
Single Unix specifications (and I think the POSIX.1g draft which adopted the
solution).  The type for a parameter describing a size is now `socklen_t', a
new type.

??kerhdr	On Linux I've got problems with the declarations in Linux
	kernel headers.

{UD,AJ} On Linux, the use of kernel headers is reduced to the minimum.  This
gives Linus the ability to change the headers more freely.  Also, user
programs are now insulated from changes in the size of kernel data
structures.

For example, the sigset_t type is 32 or 64 bits wide in the kernel.  In
glibc it is 1024 bits wide.  This guarantees that when the kernel gets a
bigger sigset_t (for POSIX.1e realtime support, say) user programs will not
have to be recompiled.  Consult the header files for more information about
the changes.

Therefore you shouldn't include Linux kernel header files directly if glibc
has defined a replacement. Otherwise you might get undefined results because
of type conflicts.

??	I don't include any kernel headers myself but the compiler
	still complains about redeclarations of types in the kernel
	headers.

{UD} The kernel headers before Linux 2.1.61 and 2.0.32 don't work correctly
with glibc.  Compiling C programs is possible in most cases but C++ programs
have (due to the change of the name lookups for `struct's) problems.  One
prominent example is `struct fd_set'.

There might be some problems left but 2.1.61/2.0.32 fix most of the known
ones.  See the BUGS file for other known problems.

??signal	Why don't signals interrupt system calls anymore?

{ZW} By default GNU libc uses the BSD semantics for signal(), unlike Linux
libc 5 which used System V semantics.  This is partially for compatibility
with other systems and partially because the BSD semantics tend to make
programming with signals easier.

There are three differences:

* BSD-style signals that occur in the middle of a system call do not
  affect the system call; System V signals cause the system call to
  fail and set errno to EINTR.

* BSD signal handlers remain installed once triggered.  System V signal
  handlers work only once, so one must reinstall them each time.

* A BSD signal is blocked during the execution of its handler.  In other
  words, a handler for SIGCHLD (for example) does not need to worry about
  being interrupted by another SIGCHLD.  It may, however, be interrupted
  by other signals.

There is general consensus that for `casual' programming with signals, the
BSD semantics are preferable.  You don't need to worry about system calls
returning EINTR, and you don't need to worry about the race conditions
associated with one-shot signal handlers.

If you are porting an old program that relies on the old semantics, you can
quickly fix the problem by changing signal() to sysv_signal() throughout.
Alternatively, define _XOPEN_SOURCE before including <signal.h>.

For new programs, the sigaction() function allows you to specify precisely
how you want your signals to behave.  All three differences listed above are
individually switchable on a per-signal basis with this function.

If all you want is for one specific signal to cause system calls to fail and
return EINTR (for example, to implement a timeout) you can do this with
siginterrupt().


??string	I've got errors compiling code that uses certain string
	functions.  Why?

{AJ} glibc 2.1 has special string functions that are faster than the normal
library functions.  Some of the functions are additionally implemented as
inline functions and others as macros.

The optimized string functions are only used when compiling with
optimizations (-O1 or higher).  The behavior can be changed with two feature
macros:

* __NO_STRING_INLINES: Don't do any string optimizations.
* __USE_STRING_INLINES: Use assembly language inline functions (might
  increase code size dramatically).

Since some of these string functions are now additionally defined as macros,
code like "char *strncpy();" doesn't work anymore (and is unnecessary, since
<string.h> has the necessary declarations).  Either change your code or
define __NO_STRING_INLINES.

{UD} Another problem in this area is that gcc still has problems on machines
with very few registers (e.g., ix86).  The inline assembler code can require
almost all the registers and the register allocator cannot always handle
this situation.

One can disable the string optimizations selectively.  Instead of writing

	cp = strcpy (foo, "lkj");

one can write

	cp = (strcpy) (foo, "lkj");

This disables the optimization for that specific call.

?? I get compiler messages "Initializer element not constant" with
	stdin/stdout/stderr. Why?

{RM,AJ} Constructs like:
static FILE *InPtr = stdin;

lead to this message.  This is correct behaviour with glibc since stdin is
not a constant expression.  Please note that a strict reading of ISO C does
not allow above constructs.

One of the advantages of this is that you can assign to stdin, stdout, and
stderr just like any other global variable (e.g. `stdout = my_stream;'),
which can be very useful with custom streams that you can write with libio
(but beware this is not necessarily portable).  The reason to implement it
this way were versioning problems with the size of the FILE structure.

To fix those programs you've got to initialize the variable at run time.
This can be done, e.g. in main, like:

static FILE *InPtr;
int main(void)
{
  InPtr = stdin;
}

or by constructors (beware this is gcc specific):

static FILE *InPtr;
static void inPtr_construct (void) __attribute__((constructor));
static void inPtr_construct (void) { InPtr = stdin; }


??	I can't compile with gcc -traditional (or
	-traditional-cpp). Why?

{AJ} glibc2 does break -traditional and -traditonal-cpp - and will continue
to do so.  For example constructs of the form:

enum {foo
#define foo foo
}

are useful for debugging purposes (you can use foo with your debugger that's
why we need the enum) and for compatibility (other systems use defines and
check with #ifdef).

??	I get some errors with `gcc -ansi'. Isn't glibc ANSI compatible?

{AJ} The GNU C library is compatible with the ANSI/ISO C standard.  If
you're using `gcc -ansi', the glibc includes which are specified in the
standard follow the standard.  The ANSI/ISO C standard defines what has to be
in the include files - and also states that nothing else should be in the
include files (btw. you can still enable additional standards with feature
flags).

The GNU C library is conforming to ANSI/ISO C - if and only if you're only
using the headers and library functions defined in the standard.

??	I can't access some functions anymore.  nm shows that they do
	exist but linking fails nevertheless.

{AJ} With the introduction of versioning in glibc 2.1 it is possible to
export only those identifiers (functions, variables) that are really needed
by application programs and by other parts of glibc.  This way a lot of
internal interfaces are now hidden.  nm will still show those identifiers
but marking them as internal.  ISO C states that identifiers beginning with
an underscore are internal to the libc.  An application program normally
shouldn't use those internal interfaces (there are exceptions,
e.g. __ivaliduser).  If a program uses these interfaces, it's broken.  These
internal interfaces might change between glibc releases or dropped
completely.

??	When using the db-2 library which comes with glibc is used in
	the Perl db modules the testsuite is not passed.  This did not
	happen with db-1, gdbm, or ndbm.

{UD} You are using an outdated copy of the DB_File Perl module.  In fact db-2
finally removed the handling of zero-sized keys which was one of the features
tested by the old Perl testsuite and therefore you see an error.  But this
never was documented and guaranteed, only broken programs used this feature.

Consequently db-2 does not need to support this feature and instead signals
an error which leads to easier debugging.  The DB_File module maintainer
Paul Marquess <pmarquess@bfsec.bt.co.uk> acknowledged this change and fixed
the testsuite so that if you use DB_File v1.60 or later you should not have
any more problems with db-2.

??	The pow() inline function I get when including <math.h> is broken.
	I get segmentation faults when I run the program.

{UD} Nope, the implementation is correct.  The problem is with egcs version
prior to 1.1.  I.e., egcs 1.0 to 1.0.3 are all broken (at least on Intel).
If you have to use this compiler you must define __NO_MATH_INLINES before
including <math.h> to prevent the inline functions from being used.  egcs 1.1
fixes the problem.  I don't know about gcc 2.8 and 2.8.1.

??	The sys/sem.h file lacks the definition of `union semun'.

{UD} Nope.  This union has to be provided by the user program.  Former glibc
versions defined this but it was an error since it does not make much sense
when thinking about it.  The standards describing the System V IPC functions
define it this way and therefore programs must be adopted.

??	Why has <netinet/ip_fw.h> disappeared?

{AJ} The corresponding Linux kernel data structures and constants are
totally different in Linux 2.0 and Linux 2.1.  This situation has to be
taken care in user programs using the firewall structures and therefore
those programs (ipfw is AFAIK the only one) should deal with this problem
themselves.


? Miscellaneous

??	After I changed configure.in I get `Autoconf version X.Y.
	or higher is required for this script'.  What can I do?

{UD} You have to get the specified autoconf version (or a later one)
from your favorite mirror of ftp.gnu.org.

??	When I try to compile code which uses IPv6 headers and
	definitions on my Linux 2.x.y system I am in trouble.
	Nothing seems to work.

{UD} The problem is that IPv6 development still has not reached a point
where the headers are stable.  There are still lots of incompatible changes
made and the libc headers have to follow.

{PB} The 2.1 release of GNU libc aims to comply with the current versions of
all the relevant standards.  The IPv6 support libraries for older Linux
systems used a different naming convention and so code written to work with
them may need to be modified.  If the standards make incompatible changes in
the future then the libc may need to change again.

IPv6 will not work with a 2.0.x kernel.  When kernel 2.2 is released it
should contain all the necessary support; until then you should use the
latest 2.1.x release you can find.  As of 98/11/26 the currently recommended 
kernel for IPv6 is 2.1.129.

Also, as of the 2.1 release the IPv6 API provided by GNU libc is not
100% complete.  In particular the getipnodebyname and getipnodebyaddr
functions are not implemented.

??	When I set the timezone by setting the TZ environment variable
	to EST5EDT things go wrong since glibc computes the wrong time
	from this information.

{UD} The problem is that people still use the braindamaged POSIX method to
select the timezone using the TZ environment variable with a format EST5EDT
or whatever.  People, read the POSIX standard, the implemented behaviour is
correct!  What you see is in fact the result of the decisions made while
POSIX.1 was created.  We've only implemented the handling of TZ this way to
be POSIX compliant.  It is not really meant to be used.

The alternative approach to handle timezones which is implemented is the
correct one to use: use the timezone database.  This avoids all the problems
the POSIX method has plus it is much easier to use.  Simply run the tzselect
shell script, answer the question and use the name printed in the end by
making a symlink to /usr/share/zoneinfo/NAME (NAME is the returned value
from tzselect) from the file /etc/localtime.  That's all.  You never again
have to worry.

So, please avoid sending bug reports about time related problems if you use
the POSIX method and you have not verified something is really broken by
reading the POSIX standards.

??	What other sources of documentation about glibc are available?

{AJ} The FSF has a page about the GNU C library at
<http://www.gnu.org/software/libc/>.  The problem data base of open and
solved bugs in GNU libc is available at
<http://www-gnats.gnu.org:8080/cgi-bin/wwwgnats.pl>.  Eric Green has written
a HowTo for converting from Linux libc5 to glibc2.  The HowTo is accessable
via the FSF page and at <http://www.imaxx.net/~thrytis/glibc>.  Frodo
Looijaard describes a different way installing glibc2 as secondary libc at
<http://huizen.dds.nl/~frodol/glibc>.

Please note that this is not a complete list.


Answers were given by:
{UD} Ulrich Drepper, <drepper@cygnus.com>
{DMT} David Mosberger-Tang, <davidm@AZStarNet.com>
{RM} Roland McGrath, <roland@gnu.org>
{AJ} Andreas Jaeger, <aj@arthur.rhein-neckar.de>
{EY} Eric Youngdale, <eric@andante.jic.com>
{PB} Phil Blundell, <Philip.Blundell@pobox.com>
{MK} Mark Kettenis, <kettenis@phys.uva.nl>
{ZW} Zack Weinberg, <zack@rabi.phys.columbia.edu>
{TK} Thorsten Kukuk, <kukuk@vt.uni-paderborn.de>
{GK} Geoffrey Keating, <geoffk@ozemail.com.au>
{HJ} H.J. Lu, <hjl@gnu.org>

Local Variables:
 mode:outline
 outline-regexp:"\\?"
  fill-column:76
End: